Singapore

Many existing financial math reasoning benchmarks suffer from data contamination and high manual construction costs. To address this, we propose a novel formula-driven approach to dynamically construct math reasoning benchmarks in finance. Our two-stage approach: (1) generates single-formula questions by LLMs using a &quot;Mask-for-Solve&quot; paradigm for ground truth answers, and (2) synthesizes multi-formula questions through hierarchical tree-based DAGs. Our approach ensures novelty (via LLMs&#39; creativity) and controllability of difficulty (via DAG structure). Based on a self-constructed financial formula bank, we utilize the proposed method to build FinMathBench, the first formula-driven and fully LLM-generated benchmark aimed at assessing LLMs&#39; math reasoning abilities in finance, containing 946 questions across 4 complexity levels. Evaluation results on 40 LLMs demonstrate significant accuracy drops in multi-formula questions, e.g., 72.9\% (1-Formula) $\rightarrow$ 14.0\% (4-Formula) for GPT-4o under Chain-of-Thought prompting. Three critical flaws of LLMs are also observed: poor direct calculation performance, bias toward frequently solved variables in formulas, and erroneous &quot;correction&quot; of valid but extreme financial values. These findings highlight gaps in current LLMs&#39; domain-specific reasoning and underscore FinMathBench&#39;s value for advancing robust financial LLMs.

AAAI 2026

FinMathBench: A Formula-Driven Benchmark for Evaluating LLMs’ Math Reasoning Capabilities in Finance

financial mathematics

formula-driven

data contamination

large language models

question generation

evaluation

[![](https://assets.underline.io/uploads/markdown_image/1/image/81a3c0317d24f663b49b996875024d45.png)](https://aclanthology.org/volumes/2022.coling-1/)



### THE CONFERENCE WE KNOW AND WE WANT.
**COLING**, the International Conference on Computational Linguistics, is one of the premier conferences for the natural language processing and computational linguistics.

First established in 1965, the biennial COLING conference is held in diverse parts of the globe and attracts participants from both top-ranked research centers and emerging countries. Today, the most important developments in our field are taking place not only in universities and academic research institutes but also in industrial research departments including tech-startups. COLING provides opportunities for all these communities to showcase their exciting discovery.

In fall of 2022, COLING will be held in Gyeongju in a hybrid format. All participants can either present at the venue site or join virtually. As more people get vaccinated, we are happy to provide safer environments for our colleagues. We believe that COLING 2022 will be one of the conferences, free from the pandemic. The hybrid format gives presenters and sponsors a valuable opportunity to promote their companies in both an online and in-person venue. For the first time in a long time, customers can interact with their sponsor's products first-hand. The online venue, too, gives sponsors the chance to network with those unable to attend the in-person session.

The hybrid format ultimately lends itself to greater exposure for our sponsors: COLING2022 will let you reach more potential partners and customers than ever before!

To gain access to this event page you are required to register. Find more information and pay the registration fee on event the organizer’s website **[https://coling2022.org/reg](https://coling2022.org/reg)**

COLING 2022

COLING, the International Conference on Computational Linguistics, is one of the premier conferences for the natural language processing and computational linguistics.

Many existing financial math reasoning benchmarks suffer from data contamination and high manual construction costs. To address this, we propose a novel formula-driven approach to dynamically construct math reasoning benchmarks in finance. Our two-stage approach: (1) generates single-formula questions by LLMs using a "Mask-for-Solve" paradigm for ground truth answers, and (2) synthesizes multi-formula questions through hierarchical tree-based DAGs. Our approach ensures novelty (via LLMs' creativity) and controllability of difficulty (via DAG structure). Based on a self-constructed financial formula bank, we utilize the proposed method to build FinMathBench, the first formula-driven and fully LLM-generated benchmark aimed at assessing LLMs' math reasoning abilities in finance, containing 946 questions across 4 complexity levels. Evaluation results on 40 LLMs demonstrate significant accuracy drops in multi-formula questions, e.g., 72.9\% (1-Formula) $\rightarrow$ 14.0\% (4-Formula) for GPT-4o under Chain-of-Thought prompting. Three critical flaws of LLMs are also observed: poor direct calculation performance, bias toward frequently solved variables in formulas, and erroneous "correction" of valid but extreme financial values. These findings highlight gaps in current LLMs' domain-specific reasoning and underscore FinMathBench's value for advancing robust financial LLMs.

poster

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

We aim to develop a goal specification method that is semantically clear, spatially sensitive, domain-agnostic, and intuitive for human users to guide agent interactions in 3D environments. Specifically, we propose a novel cross-view goal alignment framework that allows users to specify target objects using segmentation masks from their camera views rather than the agent’s observations. We highlight that behavior cloning alone fails to align the agent’s behavior with human intent when the human and agent camera views differ significantly. To address this, we introduce two auxiliary objectives: cross-view consistency loss and target visibility loss, which explicitly enhance the agent's spatial reasoning ability. According to this, we develop ROCKET-2, a state-of-the-art agent trained in Minecraft, achieving an improvement in the efficiency of inference $3\times$ to $6\times$. We demonstrate that ROCKET-2 can directly interpret goals from human camera views, enabling better human-agent interaction. Remarkably, ROCKET-2 demonstrates zero-shot generalization capabilities: despite being trained exclusively on the Minecraft dataset, it can adapt and generalize to other 3D environments like Doom, DMLab, and Unreal through a simple action space mapping.

Steering Visuomotor Policy in Open Worlds via Cross-View Goal Alignment

Vector quantization has emerged as a powerful tool in large-scale multimodal models, unifying heterogeneous representations through discrete token encoding. However, its effectiveness hinges on robust codebook design. Current prototype-based approaches relying on trainable vectors or clustered centroids fall short in representativeness and interpretability, even as multimodal alignment demonstrates its promise in vision-language models. To address these limitations, we propose a simple multimodal prompting-driven quantization framework for point cloud analysis. Our methodology is built upon two core insights: 1) Text embeddings from pre-trained models inherently encode visual semantics through many-to-one contrastive alignment, naturally serving as robust prototype priors; and 2) Multimodal prompts enable adaptive refinement of these prototypes, effectively mitigating vision-language semantic gaps. The framework introduces a dual-constrained quantization space, enforced by compactness and separation regularization, which seamlessly integrates visual and prototype features, resulting in hybrid representations that jointly encode geometric and semantic information. Furthermore, we employ Gumbel-Softmax relaxation to achieve differentiable discretization while maintaining quantization sparsity. Extensive experiments on the ModelNet40 and ScanObjectNN datasets clearly demonstrate the superior effectiveness of the proposed method.

Point Cloud Quantization Through Multimodal Prompting for 3D Understanding

Sample selection is a straightforward technique to combat noisy labels, aiming to prevent mislabeled samples from degrading the robustness of neural networks. 
However, existing methods mitigate compounding selection bias either by leveraging dual-network disagreement or additional forward propagations, leading to multiplied training overhead.
To address this challenge, we introduce $\textit{Jump-teaching}$, an efficient sample selection framework for debiased model update and simplified selection criterion. Based on a key observation that a neural network exhibits significant disagreement across different training iterations, Jump-teaching proposes a jump-manner model update strategy to enable self-correction of selection bias by harnessing temporal disagreement, eliminating the need for multi-network or multi-round training. 
Furthermore, we employ a sample-wise selection criterion building on the intra variance of a decomposed single loss for a fine-grained selection without relying on batch-wise ranking or dataset-wise modeling. 
Extensive experiments demonstrate that Jump-teaching outperforms state-of-the-art counterparts while achieving a nearly overhead-free selection procedure, which boosts training speed by up to $4.47\times$ and reduces peak memory footprint by $54\\%$.

Jump-teaching: Combating Sample Selection Bias via Temporal Disagreement

Image-based feature representation plays a critical role in visual localization, enabling robots to estimate their position and orientation in GPS-denied environments. However, this task is often undermined by significant variations in camera viewpoints and scene appearances. Recently, map-free visual relocalization (MFVR) has emerged as a promising paradigm due to its compatibility with lightweight deployment and privacy isolation on mobile devices. In this paper, we propose the Debiased Multiplex Tokenizer (DeMT) as a novel method for versatile and efficient MFVR. Specifically, DeMT performs relative pose regression through an integrated framework built upon a pretrained vision Mamba encoder, comprising three key modules: First, Multiplex Interactive Tokenization yields robust image tokens with non-local affinities and cross-domain descriptions; Second, Debiased Anchor Registration facilitates anchor token matching through proximity graph retrieval and causal pointer debiasing; Third, Orthogonal Pose Regression enhances both pair-wise and multi-view pose regression via Jacobi polynomial parsing of Kolmogorov–Arnold networks. Extensive evaluations across ten public datasets demonstrate that DeMT substantially outperforms existing benchmarks and ablation variants in diverse indoor and outdoor environments. Our code and models will be released upon paper acceptance.

Debiased Multiplex Tokenizer for Efficient Map-Free Visual Relocalization

Vision-Language Models (VLMs) such as GPT-4o now demonstrate a remarkable ability to infer users' locations from public shared images, posing a substantial risk to geoprivacy. Although adversarial perturbations offer a potential defense, current methods are ill-suited for this scenario: they often perform poorly on high-resolution images and low perturbation budgets, and may introduce irrelevant semantic content. To address these limitations, we propose GeoShield, a novel adversarial framework designed for robust geoprivacy protection in real-world scenarios. GeoShield comprises three key modules: a feature disentanglement module that separates geographical and non-geographical information, an exposure element identification module that pinpoints geo-revealing regions within an image, and a scale-adaptive enhancement module that jointly optimizes perturbations at both global and local levels to ensure effectiveness across resolutions. Extensive experiments on challenging benchmarks show that GeoShield consistently surpasses prior methods in black-box settings, achieving strong privacy protection with minimal impact on visual or semantic quality. To our knowledge, this work is the first to explore adversarial perturbations for defending against geolocation inference by advanced VLMs, providing a practical and effective solution to escalating privacy concerns.

GeoShield: Safeguarding Geolocation Privacy from Vision-Language Models via Adversarial Perturbations

Omnidirectional videos (ODVs) provide an immersive visual experience by capturing the 360$^{\circ}$ scene. With the rapid advancements in virtual/augmented reality, metaverse, and generative artificial intelligence, the demand for high-quality ODVs is surging. However, ODVs often suffer from low resolution due to their wide field of view and limitations in capturing devices and transmission bandwidth. Although video super-resolution (SR) is a capable video quality enhancement technique, the performance ceiling and practical generalization of existing methods are limited when applied to ODVs due to their unique attributes. To alleviate spatial projection distortions and temporal flickering of ODVs, we propose a Spatio-Temporal Distortion Aware Network (STDAN) with joint spatio-temporal alignment and reconstruction. Specifically, we incorporate a spatio-temporal continuous alignment (STCA) to mitigate discrete geometric artifacts in parallel with temporal alignment. Subsequently, we introduce an interlaced multi-frame reconstruction (IMFR) to enhance temporal consistency. Furthermore, we employ latitude-saliency adaptive (LSA) weights to focus on regions with higher texture complexity and human-watching interest. By exploring a spatio-temporal jointly framework and real-world viewing strategies, STDAN effectively reinforces spatio-temporal coherence on a novel ODV-SR dataset and ensures affordable computational costs. Extensive experimental results demonstrate that STDAN outperforms state-of-the-art methods in improving visual fidelity and dynamic smoothness of ODVs.

Spatio-Temporal Distortion Aware Omnidirectional Video Super-Resolution

While natural language is commonly used to guide embodied agents, the inherent ambiguity and verbosity of language often hinder the effectiveness of language-guided navigation in complex environments. To this end, we propose Visual Prompt Navigation (VPN), a novel paradigm that guides agents to navigate using only user-provided visual prompts within 2D top-view maps. This visual prompt primarily focuses on marking the visual navigation trajectory on a top-down view of a scene, offering intuitive and spatially grounded guidance without relying on language instructions. It is more friendly for non-expert users and reduces interpretive ambiguity. We build VPN tasks in both discrete and continuous navigation settings, constructing two new datasets, R2R-VP and R2R-CE-VP, by extending existing R2R and R2R-CE episodes with corresponding visual prompts. Furthermore, we introduce VPNet, a dedicated baseline network to handle the VPN tasks, with two data augmentation strategies: view-level augmentation (altering initial headings and prompt orientations) and trajectory-level augmentation (incorporating diverse trajectories from large-scale 3D scenes), to enhance navigation performance. Extensive experiments evaluate how visual prompt forms, top-view map formats, and data augmentation strategies affect the performance of visual prompt navigation.

VPN: Visual Prompt Navigation

Deep hash networks are widely used in tasks such as large-scale image retrieval due to high search efficiency and low storage costs through binary hash codes. With the growing demand for deploying deep hash networks on resource-constrained devices, it is crucial to perform network compression on them, in which automatic pruning constitutes a priority option owing to efficacy maintenance. However, most existing pruning methods are designed primarily for image classification tasks, which makes them suffer from efficacy degradation when transplanted to image retrieval tasks. In this paper, we propose a novel Automatic Channel Pruning framework by Searching with Structure Embedding (ACP-SSE). To the best of our knowledge, this is the first study to explore pruning techniques for deep hash networks and the first automatic pruning method by searching based on network topology structure. Specifically, we first design a structure encoding model by Graph Convolutional Networks (GCNs) whose graph is constructed by hash network and nodes' features are initialized by pruning strategies. The model is trained by contrastive learning loss efficiently without accuracy supervision by fine-tuning pruned models. In addition, we introduce a dynamic pruning search space in consideration of the resource constraints. By converting the automatic channel pruning task into searching the pruned structure with effect similar to the unpruned structure, it enables the method to adapt to various network architectures. Finally, the optimal networks are selected from the candidate set according to their performance in specific downstream tasks. Extensive experiments demonstrate that ACP-SSE indeed works in the automatic channel pruning area, outperforming state-of-the-art baselines in hashing-based image retrieval, while maintaining competitive accuracy in image classification. Our code is available in the supplementary material.

Automatic Channel Pruning by Searching with Structure Embedding for Hash Network

Current sparse autoencoder (SAE) approaches to neural network interpretability assume that activations can be decomposed through linear superposition into sparse, interpretable features. Despite high reconstruction fidelity, SAEs consistently fail to eliminate polysemanticity and exhibit pathological behavioral errors. We propose that neural networks encode information in two complementary spaces compressed into the same substrate: feature identity and feature integration. 
To test this dual encoding hypothesis, we develop sequential and joint-training architectures to capture identity and integration patterns simultaneously. Joint training achieves 41.3% reconstruction improvement and 51.6% reduction in KL divergence errors. This architecture spontaneously develops bimodal feature organization: orthogonal features contributing to integration pathways and the rest contributing directly to the residual. Small nonlinear components (3% of parameters) achieve 16.5% standalone improvements, demonstrating parameter-efficient capture of computational relationships crucial for behavior. Additionally, intervention experiments using 2×2 factorial stimulus designs demonstrated that integration features exhibit selective sensitivity to experimental manipulations and produce systematic behavioral effects on model outputs, including significant interaction effects across semantic dimensions. 
This work provides systematic evidence for (1) dual-encoding in neural representations, (2) meaningful nonlinearly encoded feature interactions, and (3) introduces an architectural paradigm shift from post-hoc feature analysis to integrated computational design, establishing foundations for next-generation SAEs.

Feature Integration Spaces: Joint Training Reveals Dual Encoding in Neural Network Representations

With the rapid growth of visual content in open-world environments, zero-shot hashing image retrieval (ZSHIR) has emerged to tackle the challenge of recognizing novel classes using attribute-level and semantic information. However, existing methods often rely on shallow fusion of multi-source cues (e.g., attributes, labels, and visual features) through external supervision or feature concatenation, failing to capture the underlying semantic structure in a generative way. Particularly, current bridging strategies between modalities suffer from information fragmentation and weak alignment, hindering the model's ability to fully understand complex attribute-visual relations. Moreover, subtle semantic gaps or “semantic drift” between seen and unseen classes further degrade inter-class separability and the scalability of hashing models. To address these issues, we propose a novel framework called Proxy Zero-Shot Hashing with Multimodal Fusion via Stable Diffusion (PZSH), which integrates generative modeling and contrastive learning. PZSH leverages a pre-trained Stable Diffusion (SD) model to synthesize multimodal content, and uses dual BLIP encoders to enhance semantic alignment across modalities. We further design a proxy hashing loss to enforce discriminative binary representations. Extensive experiments on benchmark datasets show that PZSH achieves state-of-the-art performance with stronger generalization to unseen classes. Our code is available in the supplementary material.

Content not yet available

Next from AAAI 2026

Steering Visuomotor Policy in Open Worlds via Cross-View Goal Alignment

Similar lecture

Weakly Supervised Formula Learner for Solving Mathematical Problems